Which BI tool should we pick for self‑service?

Pick the one your team can model in and govern. Looker excels with a strong semantic model. Superset + dbt metrics is solid and OSS. Power BI lands well in Microsoft shops. The tool matters less than enforcing a metric layer, RBAC/RLS, and performance SLOs.

How do we prevent dashboard sprawl?

Publish only certified datasets, require owners on every dashboard, and enforce a 90‑day auto‑archive for unused content. Track usage, deprecate loudly, and wire a monthly “content cleanup” job. Tie BI visibility to the `MART` schema only.

Can we do self‑service without Looker or a commercial semantic layer?

Yes. Start with dbt metrics or Cube, expose metrics via SQL/REST, and point Superset/Metabase at those models. You lose some UX polish but keep definitions centralized.

How do we control Snowflake/BigQuery costs as access grows?

Resource monitors, auto‑suspend, and warehouse sizing by persona. Cap BI warehouses with credit quotas, block raw access for BI roles, and pre‑aggregate heavy explores. Track P95 dashboard load and cost per 1k queries; optimize models when thresholds breach.

Where does AI‑generated code fit in?

Use AI to scaffold models and tests, but enforce contracts and CI. We regularly refactor AI‑generated SQL in code rescue engagements—convert ad‑hoc queries into dbt models, add tests, and attach owners. Treat AI output like junior dev code: review and harden it.

Data-engineering · Dec 2, 2025 · 9 minute read

Stop Shipping Dashboards on Sand: Building a Self‑Service Analytics Platform That Won’t Wake You at 2 a.m.

You don’t get self‑service by buying a BI license. You get it by treating data reliability like SRE treats uptime—and by wiring your stack so analysts don’t need to beg a data engineer for every column.

Back to all posts

Stop Shipping Dashboards on Sand: Building a Self‑Service Analytics Platform That Won’t Wake You at 2 a.m.

Key takeaways

Implementation checklist